Analyst(s): Mitch Ashley, Brad Shimmin, Nick Patience
Publication Date: November 16, 2026
Futurum research finds roughly 52% of organizations still in early stages of GenAI maturity. The report argues that platform engineering for GenAI, which extends GitOps, CI/CD, observability, and policy-as-code to AI workloads, is how enterprises move from pilots to production.
What Is Covered in This Article:
- Why roughly 52% of organizations remain in pre-production stages of GenAI maturity
- How Shadow AI recreates early-cloud Shadow IT risks, from unmanaged token spend to lost audit trails
- How GitOps, CI/CD, observability, and policy-as-code extend to GenAI workloads
- What platform and IT leaders should watch as GenAI moves into production
The News: The Futurum Group published From Experimentation to Execution: Platform Engineering for Scalable Generative AI, a report sponsored by Red Hat. Drawing on Futurum’s 1H 2026 AI Decision-Makers Survey and 1H 2026 Software Lifecycle Engineering Decision-Maker Study, the report finds that the GenAI bottleneck has moved from model capability to operations. It recommends that platform teams absorb GenAI as a new workload class instead of building parallel AI stacks.
Platform Engineering for GenAI Is How Enterprises Move Past the Pilot Stage
Analyst Take: Enterprise GenAI has a production problem, and its source is operational. Futurum’s 1H 2026 AI Decision-Makers Survey places roughly 52% of organizations in the awareness, experimentation, or optimization stages of GenAI maturity. Model access is widely available. What most organizations lack is the platform infrastructure to deploy, govern, and run AI systems alongside the rest of enterprise IT. Platform engineering for GenAI, more than model selection, now determines which organizations reach production.
Production Turns GenAI into a Distributed Systems Problem
Early proofs of concept were built to validate model capability, and they sat outside enterprise integration, governance, and lifecycle requirements. In production, GenAI becomes a composite system of models, retrieval, orchestration, integration, and increasingly, agentic execution. Multi-model routing has to balance cost, latency, and capability in real time. Inference at scale brings GPU capacity and cost constraints. Agents need runtime governance, including identity, sandboxing, tool-call authorization, and human oversight. Data science teams are organized for model development and experimentation, so these problems land on platform engineering.
Shadow AI Repeats the Early-Cloud Pattern
Without a shared operating model, line-of-business teams buy public model API access directly. Futurum sees the same dynamics that early-cloud Shadow IT created:
- Unmanaged token spend: Costs accumulate with no chargeback to the consuming business unit and surface as surprises at quarter close.
- Lost audit trails: Providers deprecate model versions, change pricing, or adjust safety filters without notifying the enterprise.
- Fragmented security: Data flows bypass enterprise controls for cost, security, and policy.
The report’s answer is for enterprises to move from token consumers to internal token providers, running governed Model-as-a-Service (MaaS) platforms behind a central API gateway with consistent identity, rate limits, cost attribution, and policy enforcement.
Platform Engineering for GenAI Extends Existing Disciplines
The practices platform teams already run carry over to GenAI:
- GitOps: Git becomes the source of truth for models, prompts, routing configurations, token quotas, and policies, in addition to infrastructure.
- AI delivery pipelines: CI/CD adds model and agent evaluation, security gates, and progressive delivery, with rollback paths that cover both application and model versions.
- Unified observability: Telemetry such as OpenTelemetry captures token consumption, reasoning chains, latency, and tool invocations, making token usage a FinOps signal. In Futurum’s 1H 2026 Software Lifecycle Engineering Decision-Maker Study, 37.4% of respondents rank AI observability as a top priority, and 30.9% rank AI agent observability.
- Policy-as-code governance: Automated red teaming runs before production, and runtime guardrails filter sensitive data and enforce business policy in production.
The report measures platform maturity by the ability to absorb new workloads without rebuilding the operating model. GenAI is the newest test of that maturity.
What to Watch:
- Whether enterprises consolidate public model API usage behind internal MaaS gateways, turning token spend into a line item each business unit owns.
- How quickly AI observability moves from stated priority to standard practice, with token usage tracked as a FinOps signal.
- Whether governance shifts into delivery pipelines, with automated red teaming and runtime guardrails as standard stages.
- How data locality, regulation, and inference economics shape workload placement across hybrid environments, including repatriation from public cloud.
Watch the on-demand Techstrong Learning webinar Building the AI Factory: CI/CD, GitOps, and Delivery for GenAI, where Red Hat’s Christopher Nuland, Chief AI Architect, and Natale Vinto, Technical Director, Hybrid Cloud and Enterprise AI, join Techstrong Group CEO Alan Shimel to discuss controlling Shadow AI and token spend, and building governance and token-level observability into the platform.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other Insights From Futurum:
Platform Engineering Goes Mainstream: Strategy Beats Technology
Google Agent Compute Engine Puts Cross-Cloud Governance to a Parity Test
The Hard(er) Challenge in Agent Governance Is Authorization
